IDK if it turned out one of my friends were an AI I would probably decide that that AI was a person. I’m not sure why AI and person are opposites.

They aren’t opposites, but neither is ability to pass for a human equivalent to being a person. Unsophisticated users can be fooled by currently existing chatbots. It seems pretty plausible that a sufficiently advanced chatbot-equivalent could fool you or me for extended periods of time without their underlying structure behaving in such a way that we would consider them people.

Report comment

Report comment

Just for the record, all my views/statements are based on genuine conviction, and not crimethink for its own sake.

(Bob Page voice:) Mathematically unlikely.

Report comment

Report comment

Report comment

Okay, I’ll be clearer- it is the difference between valuing interacting with other people and valuing interactions that have the same quality as interacting with other people (bonding, feedback, etc). If you value the first, you value the second, but the reverse isn’t necesarily true.

“Whatever for? What do you think would happen? I’m not seeing what the point you want to make is.”

I pointed out you could check if you need people by refering pets. You thought that people might care because pets still have minds. If you use something that doesn’t have a mind behind it to get the same stimuli, than you don’t care if there is a mind behind what you are interacting with.

It is a way to tease out the difference between the first and second values.

“Reading Scott’s posts isn’t something that appeals to my desire for human interaction anyway. But I don’t find the idea that odd some someone’s perceived utility could be lowered by them finding out that someone one has been talking to is an AI.”

Assuming they thought the individual isn’t someone they would ever meet or be able to use for any social advantage, why?

“IDK if it turned out one of my friends were an AI I would probably decide that that AI was a person. I’m not sure why AI and person are opposites.”

Because we’ve been reading less wrong and unless the AI is based upon an uploaded personality the odds of the AI interacting in good faith is poor.

Report comment

Report comment

There is a difference between “terminal value is other people” and “terminal value is interaction”; the latter can be spoofed even if you know it is being spoofed. Porn continued popularity is proof that the latter category is really common for important human instincts.

I’m not sure I’m following. What does “terminal value is other people” even mean? Besides, sex is not the same as interaction in general: I can see how somebody could not have sex as a terminal value.

Than try a Tamagotchi or one of the newer robot pets.

Whatever for? What do you think would happen? I’m not seeing what the point you want to make is.

That leads to the odd position that if you find out Scott was an AI tomorrow your utility from reading his posts drop.

Reading Scott’s posts isn’t something that appeals to my desire for human interaction anyway. But I don’t find the idea that odd some someone’s perceived utility could be lowered by them finding out that someone one has been talking to is an AI.

Report comment

Report comment

Congratulations! You are Elua! This is convenient, because it means we don’t have to build Her.

But if in truth you are not currently a benevolent superintelligent being capable of pre-specifying all of the results of Elua’s future decision-making, then perhaps let us accept something like the claim that concerned you, that “Elua will take the base human values and decide whether or not they support P, for all propositions P,” and that that’s okay. More than okay, even.

And if what you really meant to express is that Building Elua is Hard, well… Yes. Yes, it is. And that is why MIRI and FHI thank you for your support in this most difficult of endeavors.

Report comment